|
|
| pass - manually run add UDSP rule to interface A simulate the error by "lctl net_drop_add -s A -d B -i 2" do lctl ping B from A check lnetctl net show -v 3 and found send_count is increased on A as well as other interfaces, the same time, health stats dropped number increased too delete the rule by "lctl net_drop_del" and do lctl ping again send_count shows only A increased by 1 *health value didn't been noticed decrease, but developer checked lnet debug log and it showed the changing | - MR Node with Multiple interfaces
- Add UDSP rule that gives one of the local interfaces highest priority
- Send a PING
- Verify that the NID with the highest priority was used
- Send a PING
- Simulate an <error>
- PING msg should be queued on resend queue
- PING msg will be resent on a different interface
- Failed interface's health value will be decremented
- Failed interface will be placed on the recovery queue
- Eventually health score should recover so then only the preferred interface will be used (sensitivity > 0)
|
|
|
| pass - manualmanually run add UDSP on tcp1 then set that nid A in tcp1 in net_drop, the traffic will go either the good interface on tcp1 or failed. we got failed ping every other time. | - MR Node with Multiple interfaces: configure two networks, two NIDs on each network
- Add UDSP rule that gives one of the networks highest priority
- Start traffic
- Stop traffic
- Verify that NIDs on the network with the highest priority were used
- Start traffic
- Simulate an <error> (drop on one of the NIDs on the highest priority network)
- Failed interface's health value will be decremented
- Failed interface will be placed on the recovery queue
- Verify that the remaining interface on the failed network is used
- Eventually health score should recover so then only both interface on the preferred network will be used (sensitivity > 0)
|